Model order reduction for fusion components: Dynamic mode decomposition model of the magnetron injection gun

In nuclear fusion systems, the gyrotron is the source of electromagnetic waves, exploiting the electron maser instability to initialize and stabilize the fusion reaction. A simulator that predicts the system’s behaviour based on the fundamental physics is crucial for its control. This requires modelling each component by integrating the relevant physical phenomena for that component, and linking the resulting component models together. However, current models of fusion system components like the breeding blanket or the gyrotron are too computationally intensive for real-time simulations if high accuracy is sought and therefore high-fidelity and high-dimensional models are needed, as lumped approaches may not offer high enough accuracy for safety-critical applications such as monitoring and control. Reduced Order Models (ROM) can overcome this problem while maintaining accuracy. This paper focuses on the Magnetron Injection Gun (MIG), responsible for generating the electron beam by thermionic effect. It is essential to calculate the heat transfer within the MIG to predict the electron density on the emitter surface. This methodological paper studies the suitability of Dynamic Mode Decomposition (DMD) in capturing the non-linear dynamics and time-scales of the beam current emission considering a representative pulse provided by the FALCON facility. Results show the computational efficiency and capability of such technique in capturing temporal dynamics, performing spectral analysis, denoising data and forecasting features beyond the training dataset, positioning this paper as a foundation from which more complex models and ROM techniques can be considered.